ICONIC Workshop on Data Science and Crime
University of Strathclyde, December 11th, 2018
Event programme
The aim of the workshop was to bring academics and police forces across the UK together to further strengthen collaboration. The workshop had 35 participants, with 12 representatives from police forces and 23 academics with relevant interests ranging from data science, statistics, mathematical modelling, forensics and criminology. The presentations and discussions made it clear that there is a great interest from academics to get involved in understanding key law enforcement challenges that can be addressed by data science, and from police forces to use relevant techniques that academics develop. This appetite is demonstrated by examples of ongoing joint projects between academics and the police forces and by the growth of data science professionals within police forces. Two key barriers that currently inhibit collaboration were echoed throughout the day: (1) a lack of continuity (often academic projects have different timelines to those of police) and (2) the need for tangible and actionable benefits for police forces. Below, we give a summary of the talks and the discussions that followed.
Dr Theo Damoulas – University of Warwick / Alan Turing Institute
Dr Damoulas motivated interesting research questions that arise in law enforcement by his hands-on experience of working with the NYPD. As the first case study, he talked about 911 calls and how the data can be used to improve resource allocation which, at the moment, often relies solely on human expertise. His second case study demonstrated how pattern matching techniques have been used to identify serial offenders or members of the same offender group. This technique works by extracting common features among instances of crime and linking them together. An ongoing piece of work, some of which was recently published by his research group, uses the mathematical framework of Multivariate Point Processes. Applied to crime, this amounts to simultaneously modelling different crime types, rather than treating them individually. According to their results, the information contained in the cross-correlation between crime types helps improve the predictions.
Dr Damoulas taught on the Master’s Programme at New York University. By offering allocated places for police staff, this programme helps to bridge the gap between academia and policing. One of the key challenges that he came across during his collaborative projects is working with noisy data that often contain biases, such as under-reporting among certain demographics.
Police Scotland
Police Scotland was represented in the workshop by Dr Maria MacLennan, Academic Research Lead, who introduced members of the Academic Research department, based in Dalmarnock, Glasgow. As part of the Policing 2026 initiative, which lists strengthening effective partnerships as one its key tenets, the team is working with different partners such as the Scottish Institute for Policing Research. Dr MacLennan outlined the vision of the unit: establish organisational oversight of academic research, formalise research request process, enhance governance of academic research, encourage and promote evidence-based policing. A key role of the unit is to establish formal processes that will help form a fruitful relationship between academia and policing. As an example, she mentioned the work Police Scotland did in collaboration with the University of Edinburgh on stop & search data. Later in the presentation, members of the unit discussed the datasets that they have and listed some research questions they would like to tackle next. The audience appreciated the drive to formalise the processes required for effective collaboration.
West Midlands Police
Davin Parrott is a Principal Data Scientist at West Midlands Police. He presented a prototype of a system for reducing policing demand by early identification of harmful individuals. The output of such a system can then be used in assigning offender management resources, potentially in collaboration with social services. From the technical point of view, the system is a binary classification algorithm based on decision trees. The algorithm uses advanced statistical methods for applying feature selection. The features considered currently include criminal history, social graph analysis, demographic variables, and features extracted from free-text reports. Particularly of interest is the use of Latent Dirichlet Allocation for modelling free text. Exploiting information in the free text in a systematic manner is a challenge for many police forces. In the discussion that followed, a method which can quantify the uncertainty of the predictions was suggested as a possible amendment. Additionally, given the machine learning nature of the project, working more closely with criminologists to join up the two disciplines was suggested as a guiding principle in further research.
For academic collaborations, West Midlands Police works closely with the University of Warwick – supporting student projects and funding master’s course places for police staff.
Professor Jim Smith – University of Warwick / Alan Turing Institute
The talk by Professor Jim Smith focused on decision systems that can frustrate and pursue people radicalised into planning acts of extreme violence. His work has already been applied to UK’s Home Office strategy on tackling radicalisation. At the heart of the proposed tool is the blending of expert knowledge, mathematical models, and available data. The framework Prof Smith proposed can incorporate all three components using a Bayesian formalism that allows for quantifying uncertainty. During his talk, he stressed the importance of scientists working together with the domain experts and incorporating their knowledge into the tools that scientists build.
Metropolitan Police Service
Metropolitan Police Service (MPS) is the largest force in the UK with 12 Basic Command Units. Trevor Adams, who heads the Data Development Team, centred his presentation on data quality and numerous collaborations they have. One of their successful drives to improve data quality is a geocoding tool that was developed in-house. Having high accuracy location data is a key pre-requisite to modelling spatial aspects of crime. MPS is very open to collaborating with external partners as long as security & privacy concerns are addressed. Examples of collaboration include crime mapping and risk terrain modelling in collaboration with UCL, a Datathon organised together with Home Office data science team, and sharing data with local councils to design early intervention plans for troubled families. Some of the challenges when working with external partners are the difficulty of translating the output findings into operationally viable actions, not receiving any feedback after sharing the data, and conflicting timelines for the projects.
Opportunities for networking were available during the tea/coffee sessions and the lunch hour. The morning and afternoon sessions finished with open discussions. In terms of follow-on activity, in addition to future collaboration that may arise out of the face-to-face interactions, the ICONIC team have produced this report, will make presentations available on the project website (where permission has been granted by the presenters) and will keep attendees informed of future events.
Report by Jan Povala and Seppo Juhani, Imperial College London
iconicmath.org 